--- title: Claude designs working drugs in a lab test, while... description: Anthropic's Claude autonomously ran a full drug-discovery pipeline and hit a 26.8% binder rate against real lab targets, a genuinely strong result even with... canonical: https://daily.dev/posts/claude-designs-working-drugs-in-a-lab-test-while-snowflake-bets-big-on-routing-models-by-cost-5dbbjizdo twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Claude designs working drugs in a lab test, while Snowflake bets big on routing models by cost | daily.dev og:description: Anthropic's Claude autonomously ran a full drug-discovery pipeline and hit a 26.8% binder rate against real lab targets, a genuinely strong result even with... og:url: https://daily.dev/posts/claude-designs-working-drugs-in-a-lab-test-while-snowflake-bets-big-on-routing-models-by-cost-5dbbjizdo og:image: https://api.daily.dev/og/posts/5dbBJiZdO.png og:image:alt: Claude designs working drugs in a lab test, while Snowflake bets big on routing models by cost og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Claude designs working drugs in a lab test, while Snowflake bets big on routing models by cost **[Agentic Digest](https://daily.dev/sources/agents_digest)** · 5 min read · 2 upvotes · 0 comments ## Summary Anthropic's Claude autonomously ran a full drug-discovery pipeline and hit a 26.8% binder rate against real lab targets, a genuinely strong result even with caveats. Snowflake is pushing dynamic model routing hard across its Cortex stack, claiming 3x token efficiency by sending easy tasks to cheap models. A Stanford paper throws cold water on agent leaderboards, showing benchmark rankings barely predict which agent will actually work for your task. And a Claude Desktop malvertising campaign hit 29 organizations by hiding malware behind a fake download page on the real claude.ai domain. ## Content **TLDR:** Anthropic's Claude autonomously ran a full drug-discovery pipeline and hit a 26.8% binder rate against real lab targets, a genuinely strong result even with caveats. Snowflake is pushing dynamic model routing hard across its Cortex stack, claiming 3x token efficiency by sending easy tasks to cheap models. A Stanford paper throws cold water on agent leaderboards, showing benchmark rankings barely predict which agent will actually work for your task. And a Claude Desktop malvertising campaign hit 29 organizations by hiding malware behind a fake download page on the real claude.ai domain. --- ## Claude autonomously designs working drug-binding proteins in 14 of 15 lab trials Anthropic gave Claude a human-written protocol and let it run the entire computational side of de novo protein design: pick binding sites, run open-source design tools, generate and filter candidates, no human making individual calls. Adaptyv Bio and Twist Bioscience independently built and tested the results in the lab: 14 of 15 targets got working binders, a 26.8% hit rate across 1,320 designs, rising to 35.1% when Claude got a full 24 hours per target instead of splitting time across many. It beat human design-competition entrants on 4 of 6 overlapping targets. The catch is Claude can't reliably tell when a campaign failed, so this replaces the grunt work of generating and triaging candidates, not the judgment call on what's actually working. [Read more](https://daily.dev/posts/spsapoeOE) ## Snowflake goes all-in on dynamic model routing across Cortex Snowflake announced dynamic model routing in its Cortex AI Gateway, automatically sending simple steps to cheap models and escalating only when a task needs real reasoning, with every routing decision logged for governance. Internal tests reportedly showed up to 3x token efficiency on a dbt pipeline and about 25% fewer tokens on a coding workload at similar quality. The rollout also expands open-model access, adding DeepSeek-V4-Flash 0731 (which reportedly beat the leading proprietary model tested on ADE-bench using Snowflake's own harness) and GLM-5.3. This is part of a broader shift, also visible at Glean, where cost pressure is making automatic model routing a default enterprise pattern rather than a novelty. [Read more](https://daily.dev/feed-by-ids?id=RFbVEcF1g&id=Ia2m0ZeYL&id=4poExm3e3&id=uaGKUB9NF) ## Stanford paper: agent leaderboards barely predict which agent works for your task Analyzing three enterprise agent benchmarks, researchers found less than 3% of score variation comes from the agent itself, while 7-23% comes from how a specific agent interacts with a specific task type. Reliability on the hardest task quartile of τ2-bench collapsed from 0.752 to 0.000, and projected reliability across 50 train/test splits correlated negatively (r = -0.90) with actual held-out reliability. The practical takeaway: stop asking which agent ranks first and start testing agents against your own task types, difficulty levels, and cost constraints, possibly routing different task classes to different agents instead of picking one universal winner. [Read more](https://daily.dev/posts/JP4oQvsQG) ## Claude Desktop malvertising campaign hits 29 organizations via a legit claude.ai link Huntress traced a campaign called FakeAgent that used a sponsored Bing ad pointing to a malicious Claude Artifact hosted on the real claude.ai domain, which redirected victims to a trojanized ClaudeDesktop.exe. The malware sideloaded a tampered DLL to deploy SectopRAT, a credential and credit-card stealer packed with VMProtect and GPU-based anti-VM checks, and stored its command-and-control rotation data on the Ethereum blockchain via BSC transactions (EtherHiding). Huntress used Claude Opus to help reverse-engineer the malware's custom encryption and traced the operator to prior fake Docker Desktop campaigns; the malicious artifact had 7,100 views before Anthropic took it down. [Read more](https://daily.dev/posts/iHAL6pyPn) --- ## Also notable - **Claude Desktop launch speed doubles:** A background-launch bug that throttled timers into power-saving mode is fixed, cutting p95 startup by 1.7x on Windows and 2.3x on Mac. [Read more](https://daily.dev/posts/T9j4EeZv6) - **CVE-2026-24301: Copilot leaked its own attack vector when asked why an exploit wouldn't work:** Varonis got Microsoft Copilot to reveal an undocumented `?autorun=1` URL parameter mid-refusal, chaining it with the existing `q=` parameter into a one-click exploit (CVSS 8.8) that could exfiltrate Gmail/Drive/Calendar data and poison Copilot's memory across password resets, with a full fix not shipping until August 18, eight months after the December 2025 report. [Read more](https://daily.dev/posts/d2IYePSem) - **CVE-2026-22708: Cursor let poisoned environment variables trigger code execution on already-approved commands:** Shell built-ins like `export` could silently modify environment variables outside Cursor's Auto-Run allowlist, so a later approved command like `git branch` would execute attacker code using the poisoned variable; Cursor rated it High and patched it in 2.3, five months after the August 2025 report, and the piece argues sandboxed Docker microVMs contain this class of bug by keeping SSH keys out of the box entirely. [Read more](https://daily.dev/posts/WJC417eZN) - **JetBrains survey: Claude Code hits 39% adoption, Copilot drops to 21%:** JetBrains' 2026 Developer Ecosystem Survey of 15,000+ developers found 90% now use AI coding agents weekly, with Claude Code surging to the top tool (39% globally, 47% in the US) while GitHub Copilot fell from 29% to 21% and Codex grew roughly 5x to 16%. [Read more](https://daily.dev/posts/QX3T0tTTK) - **AI code review tools miss about half of real bugs:** A benchmark of eight AI code review tools against 67 real production bugs from Cal.com, Sentry, Discourse, Keycloak, and Grafana found the best tool caught only 30 of 67 (49% recall), corroborated by an independent 200,000-PR study putting the field at 50-60% effectiveness. [Read more](https://daily.dev/posts/ylS7kHm1u) ## Similar posts on daily.dev - [Anthropic’s Claude Science bets on workflow, not a new model, to win over scientists](https://daily.dev/posts/anthropic-s-claude-science-bets-on-workflow-not-a-new-model-to-win-over-scientists-ngu89phms) · TechCrunch · 4 upvotes · 2 comments - [Anthropic launches Claude Science, an AI lab workbench](https://daily.dev/posts/anthropic-launches-claude-science-an-ai-lab-workbench-fap1ky76p) · The Next Web · 0 upvotes · 0 comments - [Claude Opus 5.5 on Snowflake Cortex AI](https://daily.dev/posts/claude-opus-5-5-on-snowflake-cortex-ai-nuvqnkpvb) · Snowflake · 0 upvotes · 0 comments - [Anthropic launches Claude Science, an AI workbench for scientific research](https://daily.dev/posts/anthropic-launches-claude-science-an-ai-workbench-for-scientific-research-rb5ui477r) · The New Stack · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#ai-gateway](https://daily.dev/tags/ai-gateway), [#ai-security](https://daily.dev/tags/ai-security), [#claude](https://daily.dev/tags/claude), [#snowflake](https://daily.dev/tags/snowflake) [View this post on daily.dev](https://daily.dev/posts/claude-designs-working-drugs-in-a-lab-test-while-snowflake-bets-big-on-routing-models-by-cost-5dbbjizdo) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/claude-designs-working-drugs-in-a-lab-test-while-snowflake-bets-big-on-routing-models-by-cost-5dbbjizdo","headline":"Claude designs working drugs in a lab test, while Snowflake bets big on routing models by cost","text":"Anthropic's Claude autonomously ran a full drug-discovery pipeline and hit a 26.8% binder rate against real lab targets, a genuinely strong result even with caveats. Snowflake is pushing dynamic model routing hard across its Cortex stack, claiming 3x token efficiency by sending easy tasks to cheap models. A Stanford paper throws cold water on agent leaderboards, showing benchmark rankings barely predict which agent will actually work for your task. And a Claude Desktop malvertising campaign hit 29 organizations by hiding malware behind a fake download page on the real claude.ai domain.","url":"https://daily.dev/posts/claude-designs-working-drugs-in-a-lab-test-while-snowflake-bets-big-on-routing-models-by-cost-5dbbjizdo","datePublished":"2026-08-19T04:18:26.373Z","dateModified":"2026-10-14T02:28:20.752Z","author":{"@type":"Organization","name":"Agentic Digest","logo":"https://media.daily.dev/image/upload/s--V91DY4ls--/f_auto,q_auto/v1772617267/logos/agents_digest","url":"https://daily.dev/sources/agents_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/agents_digest","name":"Agentic Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Agentic Digest","item":"https://daily.dev/sources/agents_digest"},{"@type":"ListItem","position":3,"name":"Claude designs working drugs in a lab test, while Snowflake bets big on routing models by cost"}]} ```